Fertility and Family Statistics Branch
U.S. Census Bureau
Washington, DC 20233
Population Division Working Paper No. 33
This paper reports the general results of research undertaken by Census Bureau Staff. The views expressed are attributable to the authors and do not necessarily reflect those of the Census Bureau.
BACKGROUND: PRETESTING PROTOCOL
Relationship ItemCONCLUSIONS AND RECOMMENDATIONS
Marital Status Item
- Distributions of number of people by their relationship to householder: March 1998 Current Population Survey and the Census 2000 Dress Rehearsal
- Distributions of number of people by their relationship to householder by age: March 1998 Current Population Survey and the Census 2000 Dress Rehearsal
- Distributions of couple households by marital status and gender of partner and edit status: Census 2000 Dress Rehearsal
- Unedited distributions and percents for Census 2000 Dress Rehearsal Items 19a, 19b, and 19c
- Marital status by gender, age groups, and site from the Census 2000 Dress Rehearsal
- Item 19c. "How long has this grandparent been responsible for the(se) grandchildren?" current and recommended categories
Copy of Census 2000 Dress Rehearsal informational long form questionnaire
Population Working Papers
An analysis of data on family and household demographic characteristics collected in the Census 2000 Dress Rehearsal was conducted to identify potential content problems. The primary official memorandum series for the Census 2000 Dress Rehearsal reports on the administrative and mechanical aspects of conducting a census. Since content analyses of dress rehearsal data was not the principal focus of this memorandum series, This evaluation was prepared as a Population Division working paper. This evaluation examine the items on relationship-to-householder, grandparents-as-caregivers, and marital status using the current version of the Census Unedited File (CUF). Since this file represents the actual responses before any editing or allocation procedures are initiated, but after duplicate records are removed, the analyst can examine firsthand how the respondents record their answers and how the data appear when initially received by the Census Bureau. It should be noted that the results of these findings cannot be generalized to any area beyond the dress rehearsal sites. Simple distributions and derived statistics from this file are compared to national estimates from surveys such as the Current Population Surveys (CPS) to see if the responses provide a reasonable set or pattern of answers.
The first evaluation analyzes the relationship-to-householder item on the dress rehearsal. As expected, the patterns of responses by various relationship categories such as householders, spouses, children, and other relatives as well as nonrelatives are consistent with national distributions from the March 1998 CPS. The number of people who will have to be assigned another relationship code based on inconsistencies with their stated age is relatively low. In addition, the vast majority of questionnaires that had the other relative box checked also had write-in responses to the relationship item that would be picked up by the coding process. A procedure instituted to automatically code write-in responses for the 'other relative' category via an optical scanning reader proved very successful.
The second evaluation analyzes a series of grandparents-as-caregivers items which were actually tested for the first time in the dress rehearsal. They are included on Census 2000 because of the mandate by Congress to inquire about grandparents roles as caregivers to their grandchildren. The mandate requires Census 2000 to collect information on grandparents who serve as caregivers to their grandchild(ren) and whether that care is temporary or not. Analysis of data from a preliminary unedited data file, the Decennial Response File (DRF), and the CUF indicated that the initial response categories selected for the duration of care item were flawed (see question 19 in Appendix A). More than 60 percent of the respondents chose the open-ended category 'more than 12 months.' After comparing these findings with other sources, a more meaningful, expanded set of response categories was recommended and approved which should reduce the proportion of responses in the open-ended interval.
The final evaluation analyzes the marital status item on the dress rehearsal long form. The proportions of married, widowed, divorced, separated, and never married are within acceptable tolerances, even when cross-tabulated by age and gender. A small number of people under 15 years old who reported themselves as being married will be recoded as 'never married,' according to our edit procedures.
This year we have been afforded a unique opportunity to preview the results of the upcoming decennial census, Census 2000. Although implemented primarily as an operational test, the Census 2000 Dress Rehearsal was a unique data collection effort and useful for previewing the analytical results of Census 2000. We report the results of the review process conducted on three subject-matter areas included on the Census 2000 Dress Rehearsal form. This review is an integral part of the process that will ensure the accuracy of Census 2000 data.
We examine the relationship to householder item that is included on both the short and long forms, the series of items on grandparents caring for grandchildren from the long form, and the marital status item from the long form. (See Appendix A for questions on the Census forms.) The primary objective for the review of these items is the identification of potential problems in reporting or misunderstanding. We also examined these data to review the non-response rates, and to identify problems generated as a result of the editing and allocation procedures used for these items.
The U.S. Bureau of the Census regularly conducts reviews of its data quality; however the implementation of the Census 2000 Dress Rehearsal has provided the opportunity to "preview" the most important data collection effort undertaken by the Census Bureau each decade. Questions that are new or that have been modified are routinely pre-tested, as was the case with the relationship to householder item, which was tested in the 1995 National Content Survey.
The National Content Survey was designed to test different question formats for certain questions which were under consideration for inclusion in Census 2000. In part, based on the results of this test, the relationship responses were divided into two sections: people related to the householder and those not related to the householder (See Person 2 - Question 2 in Appendix). The related subsection includes the following: husband/wife, natural-born son/daughter, adopted son/daughter, stepson/stepdaughter, brother/sister, father/mother, grandchild, parent-in-law, son-in-law/daughter-in-law, and other relative. The other relative category has a write-in space to identify the exact relationship, for example, niece or uncle. The not related subsection has five answer categories: roomer/boarder, housemate/roommate, unmarried partner, foster child, and other non-relative. These relationship categories represent an increase in specificity over categories used in previous data collection efforts in order to better portray complex household situations and reduce the number of write-in responses. Of course, when more detailed categories are used, the quality of reporting becomes a major concern. This evaluation addresses this issue and investigates the effectiveness of the optical readers in processing forms.
Sometimes, of necessity, items are included in the decennial census without field testing as was the case with the series of items on grandparents. Fortunately the Census 2000 Dress Rehearsal provided us with an opportunity to field test these items before finalizing the questions for Census 2000.
In response to the Congressional requirement stated in the 1996 Welfare Reform Act passed in September, 1996, the Census Bureau developed three questions to identify short and long term durations of grandparent responsibility for their grandchildren. Cognitive testing was conducted by the Census Bureau in December 1996, to determine whether the wording and concepts were understandable to respondents. Ten cognitive interviews were completed, and problems with the terms "responsibility" and "duration of care" arose. The following questions were included in the dress rehearsal as a result of this testing and refinement:
19a. Does this person have any of his/her own grandchildren under the age of 18 living in this house or apartment? - YES - NO 19b. Is this grandparent currently responsible for most of the basic needs of any grandchild(ren) under the age of 18 who live(s) in this house or apartment? - YES - NO 19c. How long has this grandparent been responsible for the(se) grandchild(ren)? If the grandparent is financially responsible for more than one grandchild, answer the question for the grandchild whom the grandparent has been responsible for the longest period of time. - Less than 1 month - 1 to 6 months - 7 to 12 months - More than 12 months - Don't knowSkip pattern instructions were included to exclude people who do not have any grandchildren present or who were not responsible for the basic needs of any grandchildren who were living with them. While this seemed a reasonable structuring of these items based on results from the cognitive testing, results from other research conducted after questionnaire development indicate that the spell lengths used in the duration item were too brief.1 Fields' findings, using retrospective living arrangement histories from the National Survey of Families and Households (NSFH), indicated that substantially longer periods of grandparental care giving could be expected. A direct comparison between cross-sectional estimates from the dress rehearsal and the findings using the NSFH retrospective histories is not possible, as a population having completed childhood would have longer mean durations than a cross-section of children currently under 18. Nonetheless, there is reason to believe that the categories in the dress rehearsal will not provide an adequate range of duration intervals.
The marital status item exists much as it has for a number of years and was not tested in the National Content Survey. However, it now will be asked only on the long form in Census 2000. The analysis of this item will involve comparing the results gathered in the Census 2000 Dress Rehearsal with those patterns published in national sample surveys.
The CUF contains the raw data gathered from the Census 2000 Dress Rehearsal after an unduplication procedure. The dress rehearsal was conducted in three geographic areas to evaluate operational procedures for the decennial census. The three dress rehearsal sites are Sacramento, California; Columbia, South Carolina and 11 surrounding counties; and Menominee County in Wisconsin. Varying methods of data collection were employed across the sites from mail-out/mail-back census forms to enumerator visits.
The CUF is a data file that was created from the dress rehearsal data and is used as the input file for the edit and allocation process for both short form (100 percent) and long form (sample) items. One hundred percent items are those items that appear on all census questionnaires and for which everyone provides an answer. The sample items are those items that appear only on the long enumeration forms which will be mailed to approximately one in six households. The codes derived from write-in fields are the 'other relative' responses for the relationship item, the Hispanic origin item, and the race write-in items.
The items for the current evaluation are on both 100 percent and sample questionnaires. The relationship to the householder item is a 100 percent item. The marital status and grandparents as caregivers items are sample items. The gender and age items also utilized in this analysis are 100 percent items. As previously mentioned, 100 percent items are from all census forms, and the sample items are only from the long forms. It should be noted that the long forms contain both the 100 percent items and sample items.
Bivariate cross-tabulations provide the basic descriptive statistics for the analysis of the data for the topical areas in this report. For the relationship item, we compare the distributions of the relationship variable on the Census Unedited File (CUF) for each dress rehearsal site. Two age groups are used, 'under 15' and '15 and over.' These groups were selected, because the edits of seven relationship categories are based on a minimum age of fifteen. These seven relationship categories are 'Householder,' 'Husband/Wife,' 'Father/Mother,' 'Father-in-law/Mother-in-law,' 'Daughter-in-law/Son-in-law,' 'Housemate/Roommate,' and 'Unmarried Partner.'
The grandparents as caregivers items on the CUF are analyzed for each dress rehearsal site. For the purposes of this analysis, a preliminary age edit was performed. People under the age of 30 were not included in the analysis, because it is unlikely that people under 30 will be grandparents.2 The current edit for the Dress Rehearsal screens out people under 30 in the post-data collection edit.
The third item for review, the marital status item, is similarly evaluated on the CUF for each dress rehearsal site. Again, marital status is now a long form item only. Marital status is cross-tabulated by age category and gender to evaluate the reasonableness of the data.
Although the dress rehearsal was not designed for national comparisons, the March 1998 CPS is used as a comparison tool to evaluate the reasonableness of the relationship to the householder distributions observed in the Census 2000 Dress Rehearsal. Overall, the distributions of people in each of the relationship categories for each of the three dress rehearsal sites are consistent with the March 1998 CPS, as shown in Table 1. Between 0.75 and 1.26 percent of the dress rehearsal responses are coded as blank or unreadable. Less than three percent selected the 'other relative' category which is lower than in the CPS because of the inclusion of various step/adopted child and in-law categories in the dress rehearsal. Those who selected the 'other relative' category had a write-in response rate of about 76 percent for Sacramento, 94 percent for Columbia, and only 43 percent for Menominee. However, over half of all the write-ins recorded were unnecessary. Respondents provided valid marks in pre-specified categories but also wrote in descriptions in the write-in boxes. Finally, Table 1 shows that there are slightly higher proportions of Person 1 categories in the Sacramento site and the South Carolina site, possibly due to the inclusion of continuation forms where Person 6 in the household on a continuation form was counted as Person 1. This problem will be fixed in the actual processing.
Table 2 examines the relationship categories further by separating the sample into two age groups, 15 and over and under 15 years of age. A problem that became apparent immediately was that some people under 15 years of age were reported as the householder, spouse, father, mother, father/mother-in-law, daughter/son-in-law, housemate/roommate, or unmarried partner (between one and two percent of the under 15 population). These numbers should have been zero or close to zero because they represented impossible relationships. Despite these anomalies (which will be edited in the post-collection phase), the relationship distributions in these age categories were comparable to the CPS. As expected the vast majority of people under 15 are natural-born children. There are more step than adopted children, and the proportion of other nonrelatives is very small. The second most frequently identified category is grandchild. Less than two percent of children under 15 were recorded as other relatives and most of them will eventually be coded back into precoded categories or as nephews/nieces and cousins according to their write-in responses.
The 15 years and older group in Table 2 shows a slightly higher percentage of spouses in South Carolina than in Sacramento. The other relative relationship category represents about one percent of the sample. The unmarried partner category percentage in Sacramento (3.5%) is nearly twice that of South Carolina (2.0%).3 Overall, these distributions look reasonable. After processing and assigning write-ins to precoded categories and the additional write-in derived categories, less than about 0.5 percent will remain to be coded 'other relatives.'4 This confirms the validity of the decision to forego hand coding and to only use the precoded variables on the relationship data dictionary file.
Table 3 shows the distributions of couple households by marital status and gender of partner. The ratio of unmarried-partner households to married-couple households was higher in Sacramento than in South Carolina, as expected. About 16 to 20 percent of married-couple households have female householders, again comparable to CPS estimates which indicate that 23 percent of married-couple households in March 1998, had female householders. Also, a higher proportion of unmarried-partner households (50-52%) have female householders in the Dress Rehearsal sites.
The current editing process in the dress rehearsal and the anticipated editing specifications for Census 2000 do not accept relationship responses that would result in same gender married-couple households. The person reported to be the "spouse" of the householder would have his/her relationship changed to an "unmarried partner."
The editing of the relationship codes for married persons of the same gender into unmarried partners will have considerable effect on the latter estimate (Table 3, lower panel - It is important to note that these percentages are volatile because the bases they are calculated on are relatively small). In the dress rehearsal the edit would have resulted in an 8 percent increase in the aggregate number of unmarried partner households in Sacramento, a 16 percent increase in South Carolina, and a 6 percent increase in Menominee. Based on an examination of other characteristics (e.g. first names) a small proportion of these same gender married couples will have the gender of one of the spouses changed instead of the relationship of the spouse.
The effect of the current household relationship edit on the number of same gender unmarried partner households would be a 74 percent increase in Sacramento and would more than triple the number in South Carolina. This is a warning flag that the current edit specifications will have a major impact on published data for unmarried partners. The current cases in the state and federal Supreme Courts on the recognition of same gender marriages could alter the edit specifications and significantly change the nature of the edited data (see second panel of data in Table 3 for a comparison of distributions before and after incorporating current edit specifications which transforms same gender unmarried partners).
As shown in Table 2, the second most reported relationship after biological child for people under 15 years of age was grandchild. The 104th Congress recognized the importance of examining grandparent-grandchild relationships and passed legislation requiring further investigation of the subject. The 1996 Welfare Reform Act mandates the "[Census] Bureau to collect statistically significant data, in connection with its decennial census...concerning the [upward] trend [in the numbers] of grandparents who are the primary caregivers for their grandchildren." Therefore, the three grandparents as caregivers items were included on the dress rehearsal and with some modification will appear on the decennial census.
The unedited and unweighted distributions for the grandparents as caregivers items from the first returns are presented in Table 4. As expected, based on rough estimates from the CPS, about five percent of people reported that they are living with their grandchildren. Item 19b indicates that about one-half of those grandparents are responsible for their grandchildren. The distribution of the durations reported in Item 19c clearly shows that the majority of grandparents who are responsible for the care of their grandchild(ren) have been responsible for them for more than one year; 65% for Sacramento, and 70% for the South Carolina site.5 The proportions of people caring for their grandchildren for less than one month, and those who respond that they "don't know" the duration are relatively small.
Marital Status Item
Like the grandparents as caregivers item, marital status is a sample item that appears only on the census long forms. Tables 5, 6, and 7 show the marital status of the respondents by dress rehearsal site, gender, and age. First examining the data for Sacramento in Table 5, we see that the non-response rate for the marital status item is about 9 percent for people under 15 and 6 percent for people 15 years and older. Among people under 15, while the vast majority are never married, about 2 percent of those reporting on marital status are identified as either currently or formerly married. The post-edit processing procedures, however, will recode all people under 15 years as being never married as in the Dress Rehearsal.
Among people in the Sacramento dress rehearsal site 15 years and over who reported on their marital status, over 50 percent of males and 45 percent of females are "now married." The proportion now married peaks in the age groups 45 to 64 years and 65 to 74 years and then declines as the onset of widowhood increases with age. Also as expected, the proportion of people divorced peaks in the 45 to 64 year old age group. The proportion of people who have never married fell throughout the age groups, declining to 5 percent among people 75 years old and over.
The South Carolina dress rehearsal site's distribution of marital status is also shown in Table 5. The non-response rate for South Carolina was 8 percent for people under 15 years and 12 percent for those 15 years and over. Similar to the results from the Sacramento site, almost 2 percent of the people under 15 years of age in the South Carolina site selected a response other than 'never married.' Among those reporting marital status almost 60 percent of males and nearly 50 percent of females 15 years and over are 'now married.' An increase in the proportion of widowed people across the age categories is evident. There is also an increase and then a decrease with age in the proportions 'divorced' and 'separated,' similar to the pattern in Sacramento.
Table 5 shows the distribution for the Menominee, Wisconsin dress rehearsal site. The results are similar to the previous two sites. However, because the small number of total responses, a detailed analysis of these results would not produce a meaningful evaluation.
Reporting of the relationship of household members to the householder proved to be very consistent with previous administrations of this item. We find very little difference between the distributions from the Census 2000 Dress Rehearsal and the March 1998 CPS. The Dress Rehearsal results were comparable with the findings in those geographic areas from the 1990 census.
Review of the items included on the Census 2000 Dress Rehearsal to measure the prevalence, nature, and duration of grandparent-grandchild co-residence uncovered serious flaws in the categorization of the duration of co-residence item between grandchildren and the grandparents responsible for them. Without the benefit of a field test in the National Content Survey, and with only cursory cognitive testing on a handful of local grandparents, the Census 2000 Dress Rehearsal provided our best tool for substantively evaluating these questions.
In the National Survey of Families and Households (NSFH), the residential history of children living with grandparents is retrospectively collected for the period when the respondents were under 19. From an examination of these data, we found that children most often begin co-residence with their grandparents at a very young age, generally under two years old. For the period from birth to age 17 the median duration of time spent with grandparents was between three and four years. We expect this to be somewhat lower in Census 2000, as it will capture a "snapshot" of the population under 18 living with their grandparents. Younger children will not have had the opportunity to complete the time spent in the care of their grandparents. However, the fact that two-thirds of the responses to the duration of responsibility item in the dress rehearsal were recorded in the open-ended top category strongly suggests that the initial duration of care categories fail to provide an adequate portrait of grandparent-grandchild relationships.
Based on the results of our evaluation of the Census 2000 Dress Rehearsal, and a reviww of the age pattern of grandparent-grandchild co-residential experiences from the NSFH, we recommended a change in the wording of the response categories for Item 19c (see Table 6). We recommended combining the 'less than one month' and 'one to six months' categories, and that the 'don't know' category be replaced by a data category. These changes will enable a broader range of experience to be observed which more closely reflects the distributions of co-residence durations in the population and at the same time would not impact on the format size constraints for this item. Without this adjustment, this census item would yield results which have limited value, and would not reasonably respond to the Congressional requirement in the 1996 Welfare Reform Act to identify short and long term durations of grandparent responsibility for their grandchildren. These recommendations were reviewed and approved for inclusion on the Census 2000 long form.